This study addresses the challenge of segmenting strawberry leaves and lesions in natural backgrounds, which is critical for accurate disease severity assessment and automated dosing. Focusing on strawberry powdery mildew, we propose an enhanced YOLOv8-based segmentation method for leaf and lesion detection. Four instance segmentation models (SOLOv2, YOLACT, YOLOv7-seg, and YOLOv8-seg) were compared, using YOLOv8-seg as the baseline. To improve performance, SCDown and PSA modules were integrated into the backbone to reduce redundancy, decrease computational load, and enhance detection of small objects and complex backgrounds. In the neck, the C2f module was replaced with the C2fCIB module, and the SimAM attention mechanism was incorporated to improve target differentiation and reduce noise interference. The loss function combined CIOU with MPDIOU to enhance adaptability in challenging scenarios. Ablation experiments demonstrated a segmentation accuracy of 92%, recall of 85.2%, and mean average precision (mAP) of 90.4%, surpassing the YOLOv8-seg baseline by 4%, 2.9%, and 4%, respectively. Compared to SOLOv2, YOLACT, and YOLOv7-seg, the improved model’s mAP increased by 14.8%, 5.8%, and 3.9%, respectively. The improved model reduces missed detections and enhances target localization, providing theoretical support for subsequent applications in intelligent, dosage-based disease management.

This study adopted hyperspectral imaging technology combined with machine learning to detect the disease severity of stem blight through the canopy of asparagus mother stem. Several regions of interest were selected from each hyperspectral image, and the reflection spectra of the regions of interest were extracted. There were 503 sets of hyperspectral data in the training set and 167 sets of hyperspectral data in the test set. The data were preprocessed using various methods and the dimension was reduced using PCA. K−nearest neighbours (KNN), decision tree (DT), BP neural network (BPNN), and extreme learning machine (ELM) were used to establish a classification model of asparagus stem blight. The optimal model depended on the preprocessing methods used. When modeling was based on the ELM method, the disease grade discrimination effect of the FD−MSC−ELM model was the best with an accuracy (ACC) of 1.000, a precision (PREC) of 1.000, a recall (REC) of 1.000, an F1-score (F1S) of 1.000, and a norm of the absolute error (NAE) of 0.000, respectively; when the modeling was based on the BPNN method, the discrimination effect of the FD−SNV−BPNN model was the best with an ACC of 0.976, a PREC of 0.975, a REC of 0.978, a F1S of 0.976, and a mean square error (MSE) of 0.072, respectively. The results showed that hyperspectral imaging of the asparagus mother stem canopy combined with machine learning methods could be used to grade and detect stem blight in asparagus mother stems.

Fusarium wilt of banana is a soil-borne vascular disease caused by Fusarium oxysporum f. sp. cubense. The rapid and accurate detection of this disease is of great significance to controlling its spread. The research objective was to explore rapid banana Fusarium wilt latency and onset detection methods and establish a disease severity grading model. Visible/near-infrared spectroscopy analysis combined with machine learning methods were used for the rapid in vivo detection of banana Fusarium wilt. A portable visible/near-infrared spectrum acquisition system was constructed to collect the spectra data of banana Fusarium wilt leaves representing five different disease grades, totaling 106 leaf samples which were randomly divided into a training set with 80 samples and a test set with 26 samples. Different data preprocessing methods were utilized, and Fisher discriminant analysis (FDA), an extreme learning machine (ELM), and a one-dimensional convolutional neural network (1D-CNN) were used to establish the classification models of the disease grades. The classification accuracies of the FDA, ELM, and 1D-CNN models reached 0.891, 0.989, and 0.904, respectively. The results showed that the proposed visible/near infrared spectroscopy detection method could realize the detection of the incubation period of banana Fusarium wilt and the classification of the disease severity and could be a favorable tool for the field diagnosis of banana Fusarium wilt.

Modern weeding technologies include chemical weeding, non-contact methods such as laser weeding, and conventional mechanical inter-row cultivation characterized by soil loosening and weed uprooting. For maize, mechanical inter-row cultivation is key to cutting herbicide use and enhancing the soil–crop environment. This study developed a vision-guided intelligent inter-row cultivator with electric lateral shifting—its frame fabricated from Q235 low-carbon structural steel and assembled mainly via bolted and pinned joints—that computes real-time lateral deviation between the implement and crop rows through maize plant recognition and crop row fitting and uses delay compensation to command a servo-electric cylinder for precise ±15 cm inter-row adjustments corresponding to 30% of the 50 cm row spacing. To test the system’s dynamic response, 1–15 cm-commanded lateral displacements were evaluated at 0.31, 0.42, and 0.51 m/s to characterize the time-displacement response of the servo-electric shift mechanism; field tests were conducted at 0.51 m/s with three 30 m passes per maize growth stage to collect row-guidance error and root-injury data. Field results show that at an initial offset of 5 cm, the mean absolute error is 0.76–1.03 cm, and at 15 cm, the 95th percentile error is 7.5 cm. A root damage quantification method based on geometric overlap arc length was established, with rates rising with crop growth: 0.12% at the V2 to V3 stage, 1.46% at the V4 to V5 stage, and 9.61% at the V6 to V8 stage, making the V4 to V5 stage the optimal operating window. Compared with chemical weeding, the system requires no herbicide application, avoiding issues related to residues, drift, and resistance management. Compared with laser weeding, which requires high tool power density and has limited effective width, the tractor–implement system enables full-width weeding and shallow inter-row tillage in one pass, facilitating integration with existing mechanized operations. These results, obtained at a single forward speed of 0.51 m/s in one field and implement configuration, still require validation under higher speeds and broader field conditions; within this scope they support improving the precision of maize mechanical inter-row cultivation.

This study addresses the challenge of scheduling coordination between harvesters and transfer vehicles during the mechanized harvesting of processing peppers. An online detection method for assessing the loading capacity during harvesting was proposed, employing non-contact distance sensors to measure the stacking height of peppers in the hopper in real time. This measurement was used to calculate the loading capacity of the transfer vehicle for peppers. This study compared and analyzed the detection accuracy of ultrasonic, infrared distance, and light detection and ranging sensors to identify the most suitable sensor for detecting the stacking height of peppers, and establish an optimal detection model for the loading capacity of peppers in transfer vehicles. The results indicated that ultrasonic sensors achieved the highest accuracy, with a maximum absolute error of 11.0 mm and a standard deviation of 8.5 mm in detecting the stacking height of peppers. When three symmetrically mounted ultrasonic sensors were used, the maximum error and standard deviation for the mean stacking height under varying lifting speeds were 37.0 mm and 15.5 mm, respectively. The developed model’s relative error in detecting the pepper loading capacity decreased to less than 2% when the fed volume exceeded 0.4 m³. These findings provide a basis for scheduling decisions in the coordination of pepper harvesters and transfer vehicles.

Issues of poor real-time performance and low accuracy in the detection of load volume in the hopper during the mechanized harvesting of wine grapes are addressed in this study through the development of a proposed volume detection method based on ultrasonic sensors. First, the ultrasonic sensor beamwidth and detection height were determined through calibration tests. Next, a test bench was used to explore the influence of the number of ultrasonic sensors and conveying speed on the detected grape pile height. Data-based regression and hopper configuration-based geometric models correlating grape load volume with detected pile height were subsequently constructed; their accuracies were compared using test bench experiments to identify the optimal detection scheme. The regression model was more accurate than the geometric model under the considered conveying speeds with a maximum relative error of 8.0% for the former. Finally, field tests determined that the average grape load volume detection error during actual harvesting was 14.4%. Therefore, this study provides an effective solution for the detection of grape load volume in the hopper during mechanized harvesting and establishes a theoretical basis for the development of intelligent grape harvesting methods.